Oracle just signed a deal worth around $300 billion to host OpenAI's infrastructure. Three hundred billion. For a company that's never made a profit and burns cash faster than a crypto exchange in 2022.
I've been working with Oracle databases since before AWS existed. Their bread and butter has always been selling overpriced enterprise licenses to Fortune 500 companies who were too scared to switch. Now they're betting everything on hosting chatbots.
Larry Ellison thinks this'll generate $500 billion in AI revenue. That's five times what Apple makes in a year. Apple - the company that sells $1,500 phones to people who already have $1,200 phones. Oracle expects to make 5x that hosting servers for a company that's never turned a profit.
I spent three months integrating GPT-4 into a customer portal last year. You know what I learned? These models are confident idiots. They'll tell you React doesn't support hooks, then explain how to fix it using hooks. They'll generate SQL that would DELETE FROM users WHERE 1=1 and act like they solved your problem.
But apparently that's worth $300 billion in infrastructure investment.
OpenAI burns through money like it's going out of style. They're training models that hallucinate, can't do basic math without checking their work, and need constant human babysitting. I've seen startups with better unit economics selling overpriced coffee.
Oracle's stock jumped 42% when they announced this deal. Forty-two percent. Because investors think AI infrastructure is magic money printing. I remember when people said the same thing about blockchain hosting and IoT platforms.
Here's the thing - most AI projects I've worked on end up costing more than they save. MIT found that less than 10% of AI pilots actually make money. The rest are expensive solutions looking for problems that already have cheaper solutions.
Even Sam Altman said investors are "overexcited about AI." When the guy making billions off the hype tells you to calm down, maybe fucking listen.
This deal is Oracle admitting their database business is dead. They're pivoting to AI infrastructure because it's the only growth story that doesn't involve selling forty-year-old database software to companies that are too scared to migrate.
Problem is, they're betting everything on one customer who might implode next quarter. OpenAI needs to raise billions continuously just to keep the lights on. Eventually, investors will stop throwing money at a company that burns cash to train chatbots that can't reliably tell you what day it is.
When that happens, Oracle's left with $300 billion worth of servers running nothing but scheduled backup jobs and Ellison's ego.
I've seen this movie before. In 2000, companies spent trillions building fiber optic networks for demand that didn't exist. Whole telecoms went bankrupt laying cable to nowhere. At least fiber actually moved data.
Today's version: building data centers to train AI models that need humans to check their work before they accidentally order 10,000 units instead of 10. At least the railroad bubble gave us actual trains.
This deal will either make Oracle look brilliant or destroy them completely. If OpenAI somehow makes money from their chatbots, Ellison's a visionary. If not, he just signed the most expensive suicide note in corporate history.
I'm betting on the suicide note. Three hundred billion dollars for hosting services is insane when AWS exists and doesn't require you to bet your entire company on one customer who's never made a profit.